Antarctic Fiber Optic Spectrometer

We chose a Newtonian configuration as a good compromise between possible telescopes types. A prime focus system would have been too long, and a Cassegrain telescope unnecessarily complex and expensive. The telescope has been built with the supporting mount specifications in mind and, in particular, a maximum allowed swing radius of 600 mm and weight of 80 kg.

The primary parabolic mirror has a diameter of 318 mm and a focal ratio of 3.35. It is made from Astrositall, a glass with a low thermal expansion coefficient¹ of α = 0.15 × 10^-6 K⁻¹. It was given a UV‐enhanced aluminum coating of Al‐MgF₂ (yielding 85% reflectivity from 300 to 850 nm). The secondary is an elliptical flat mirror at 45° with a major axis of 108 mm and produces 6% obscuration. It is a commercial Pyrex product (α = 3.2 × 10^-6 K⁻¹) with a UV‐enhanced coating. The telescope tube is closed by a front window (which is also the aperture stop) with a diameter of 290 mm. It is 10 mm thick and is made from fused silica with a transmission of 92% over the range 300–850 nm; an antireflection coating was not used.

The telescope is designed neither to require nor to offer any degree of adjustment, thus avoiding any long‐term instability. Careful manufacturing and the use of shims under the primary mirror were used to align all the optics. The primary mirror is held in a standard cell with three fixed points of contact, three lateral supports (one is removable), and three retaining pads on the mirror surface (spring plungers). After several disappointing tests of glues (which, although rated for low‐temperature applications, proved inadequate), the secondary mirror was simply glued with RTV silicone directly onto the spider subassembly. It is worth noting that RTV silicone becomes quite hard at temperatures below −30°C but does not become brittle and so is a suitable glue for many low‐temperature applications. The entrance window was compressed onto an O‐ring by an external holder.

A critical requirement of the telescope was its ability to remain focused and collimated over a temperature ranging from + 20°C to −90°C. An athermalized design was therefore essential. We chose materials with very low thermal expansion coefficients. All the mechanical parts (including the fiber optics injection module) were made from Invar 36 (α = 0.9 × 10^-6 K⁻¹). A further advantage of this method of athermalization is that the differential thermal expansion among the materials used (Invar, Astrositall, Pyrex) is negligible and does not create any significant mechanical stress on the optics.

We also investigated the alternative approach of using an all‐aluminum design for the mechanical and optical components. This would have resulted in a light and easy‐to‐machine structure, but the preparation of aluminum mirrors for UV applications is difficult and expensive. An all‐Pyrex design was also considered, but a large Pyrex telescope tube would be too heavy for the mount.

The tube (Fig. 2) was made from a rolled Invar sheet (1.6 mm thick), which was TIG welded. A closed square box section holds it in the middle and presents an interface to bolt the telescope onto the mount. External reinforcing webs were welded longitudinally to the tube to increase stiffness. Two flanges were bolted at each extremity of the tube, the bottom one for locating the primary mirror support plate, and the upper one for mounting the window holder. Finally, the structure was stress relieved prior to final machining and optical alignment by heat treatment for 1 hr at 400°C followed by air cooling.

Zoom In Zoom Out Reset image size

The small field of view (100 μm ≈ 20 ^'') of the optical fibers at the focus of the telescope makes the design of the spider critical. This structure must be rigid enough to avoid bending or twisting with gravity. The spider (four arms, each 6 mm wide) and secondary mirror holder were cut as one single piece from an Invar plate (7 mm thick). A water‐jet cutter was used in order to avoid introducing stress. NASTRAN modeling of the telescope indicated that the maximum deflection of the tube, in going from the zenith to the horizon, creates a displacement of the optical axis of only 15 μm in the focal plane.

A major problem with all instruments operated in the cold is frosting of the optical surfaces. This is a constant challenge, and it requires a lot of care to implement an effective solution. Good practices include careful sealing using O‐rings on all removable parts, then flushing the inside volume with High Purity Grade dry nitrogen (which nevertheless still has 10⁻⁴% water content) before sealing (with a small positive pressure in the volume), and using an internal desiccant to remove any remaining water vapor (at least to a final partial pressure of water lower than the vapor pressure of ice at −90°C). If this process is not perfect, the remaining water will freeze on the coldest surface, which, in a telescope, is usually the entrance window.

We used fluorosilicone O‐rings (which stay flexible at low temperatures) and a canister filled with about 50 g of finely ground calcium hydride (CaH₂), attached to the AFOS telescope tube with a standard vacuum KF flange. CaH₂ was chosen because the equilibrium water vapor pressure above it is only 10⁻⁵ mm Hg, well below the 7 × 10^-5 mm Hg vapor pressure of ice at −90°C (Schriver 1969, p. 194). The CaH₂ powder was placed behind a Gortex membrane to allow water vapor to pass and to keep the powder in one place. Reaction of CaH₂ with the residual water vapor generates hydrogen gas. However, the quantities involved in this application are very small and should not create problems.

"Diamond dust," composed of micron‐sized cylindrical crystals of ice, is a common phenomenon in Antarctica. It can penetrate any small gaps and rapidly fill large volumes. To prevent this problem, and more generally to avoid accumulation of snow, all the external surfaces of the telescope must be as flat and smooth as possible and properly sealed.

Finally, we used a Brownell bidirectional pressure relief valve, set to open at a differential of 0.15 psi (at −80°C), in order to prevent dangerous pressures from building up. This allows some "breathing" of the telescope air space, which effectively increases the volume of air that the CaH₂ desiccant has to process during the year. We calculated the importance of this effect by using historical atmospheric pressure/temperature data from the 1995 South Pole records. As Figure 3 shows, this additional volume of air is small enough that the desiccant should easily be able to cope.

Zoom In Zoom Out Reset image size

We have designed a system that allows the detection of ice formation on the entrance window and removes it if necessary. The ice sensor is formed by a ring of four LEDs in series, attached to the internal walls of the tube and illuminating the window from an oblique angle. Any formation of ice will create backscattering of the light into the optical system and will produce increased signal onto the detector. By comparing with a stored reference signal from a clear window, we can determine if the deicing system needs to be activated or if astronomical observations can continue.

After considering many possibilities, it appeared that the best approach to deice the entrance window was a heated aluminum plate placed in close proximity to the glass. Usually, a temperature differential of 5°C–10°C above ambient is sufficient to sublimate ice.

The heating cover is attached to the mount platform on a pivot axis and is balanced by a counterweight. When the telescope is parked and pointing toward the nadir, the front surface of the window is brought within 1 mm of the heater. The heater consists of 12 resistors, capable of dissipating a total of 20 W, bolted to a 6 mm thick aluminum plate insulated on the side away from the window.

A platinum temperature sensor is glued with RTV silicone to the center of the window on the inside surface. The window itself is mounted on a thin‐walled stainless steel structure welded to the top flange of the Invar tube. This was done in order to create a large thermal impedance and to minimize transfer of heat from the window to the Invar structure.

The window itself has a step machined around the edge and is compressed onto an O‐ring by a retaining ring. The resulting external surface is flat, and a ring of felt has been inserted between the glass and the holder to break the thermal path and to seal against ice crystals.

A special UV‐enhanced fiber (high OH content) is necessary to achieve a reasonable transmission in the UV. These fibers have a deep OH absorption band at 730 nm. We would degrade the performance of our instrument in the red if we used only this fiber. In particular, we would have had to increase the integration time (as we need a very high signal‐to‐noise ratio) for the measurements of atmospheric H₂O between 700 and 800 nm.

A second problem arises from the necessity to filter out second‐order light at wavelengths below 400 nm when observing between 600 and 800 nm. With a single fiber we would be unable to observe simultaneously in the blue and red regions of the spectrum.

Hence, we decided to use two types of fibers (see Table 2): a "blue" one (FV series from Polymicro Technologies 1994) and a "red" one (FI series from Polymicro Technologies 1994, ultra–low OH, without absorption in the red but with poor UV transmission). A dichroic beamsplitter (ideally with R = 100% from 300–550 nm and T = 100% from 550–850 nm) directs the light to the respective fibers. The separation at 550 nm is chosen because this is the wavelength at which both the red and blue fibers reach the same transmission (and there are no special observational requirements at this wavelength). This concept allows us to optimize the transmission from 300 to 850 nm and to make simultaneous observations over the entire spectrum, which gives an effective gain in observing efficiency.

A 100 μm fiber core corresponds to a 196 of field of view (FOV) in the sky and 2.4 nm FWHM spectral resolution on the CCD. We chose it because, among the available fibers, that was our best compromise between decreasing core diameter resulting in difficulties centering the star image at the input of the fiber and putting more severe constraints on the 20 ^'' pointing accuracy of the mount, and increasing core diameter and losing spectral resolution (as the fibers themselves act as the entrance slit to the spectrometer).

The optical fibers we used were a step‐index silica type with a standard numerical aperture of 0.22 ± 0.02.

We were concerned by two problems: minimizing focal ratio degradation (or FRD, caused by irregularities or microbends inside the fiber, scattering the light and decreasing the focal ratio of the guided beam) and achieving a good performance at low temperature. The first issue is solved by having a fast input beam (f/3–f/4) and choosing a high core/cladding ratio of 1.4. This choice does not affect the low‐temperature performance, for which a relatively thin buffer made of polyimide is recommended (for silica: α = 0.55 × 10^-6 K⁻¹ and for polyimide: α = 0.6–1 × 10^-6 K⁻¹).

The composition finally chosen was 100/140/170 μm (±7/5/8) of silica/doped silica/polyimide.

In order to subtract the sky brightness from the star spectrum, we observe the star and the sky with separate fibers. Two sky fibers are used to increase our precision and have a wider sky coverage for pointing purposes; i.e., there are two bundles (a red one and a blue one), with three fibers in each. The two sets of fibers are set at 90° from each other in an L‐shaped configuration.

Finally, we put each bundle of fibers in a Teflon tube (inside diameter = 1.07 mm, outside diameter = 1.89 mm) in order to protect them. The Teflon was etched in a hydrofluoric bath to facilitate the adhesion of glue. We required a cable design that did not stress the fibers through differential thermal expansion (hence the fibers are loose inside the tubing) and that remains flexible at low temperatures. The sections that were more exposed to manipulation on the telescope mount and inside the shelter were protected by another Teflon tube (inside diameter = 4.22 mm, outside diameter = 5.24 mm) that encloses the two bundles.

To avoid potential failure points and to simplify the system, we decided to use a single continuous fiber between the telescope and the spectrograph. Having several segments connected together would be feasible by using appropriate connectors and choosing a refractive index matching fluid that remains transparent at low temperature, but this introduces potential unreliability.

Watson & Terry (1995) have shown that for a star diameter far smaller than the fiber core (in our case, we have a ratio of ≈20), the radial scrambling in the fiber is poor, and the output intensity profile is not flat but centrally peaked. If the star is not centered on the fiber core, the output intensity profile becomes annular. The radius of this illuminated ring is equal to the offset of the star spot from the fiber center. Further investigations will have to be conducted to determine if this effect will adversely affect our data.

This module (Fig. 4) is mounted on a flange at the Newtonian focus at the side of the telescope. It is positioned at 45° from the telescope spider arms to avoid light from diffraction spikes entering the optical fibers. The module consists of a small Invar block (≈80 × 50 × 50 mm) incorporating a beamsplitter that directs blue and red light into the appropriate fiber bundle. Every component of the module is sealed with O‐rings where necessary and is attached with dowel pins to allow precise and reproducible positioning.

Zoom In Zoom Out Reset image size

After preliminary optical assembly of the telescope tube, the position of the focus relative to the focal plane flange was determined by a Foucault test and was found to be only 500 μm off the design position. Following this measurement, the final longitudinal machining of the injection module was carried out to bring the fibers into perfect focus.

A dichroic beamsplitter (BK7 substrate, 10 mm square, 1 mm thick) was made by deposition of a 31 layer coating onto the first surface (R. C. Schaeffer 1996, private communication from A. G. Thompson and Co. Pty. Ltd. [Australia]). We chose a 20° incident angle to minimize the difference in cutoff wavelength between the two polarizations and to reduce astigmatism in the transmitted beam. The disadvantage of this acute angle is an awkward and much reduced space to connect the fibers that are close to the incoming beam. The slope of the cutoff obtained was less steep than expected, with the transmission ranging from 15% at 545 nm to 85% at 656 nm. Its position was also slightly off the specified value with a transmission/reflection ratio of 50% at 592 nm. The glass is glued onto a holder that is inserted into the Invar module from a hole on the side. Ray‐tracing confirmed that astigmatism generated by the glass in the transmitted beam was small (rms and geometric spot sizes were 6.4 and 14 μm, respectively) at 850 nm.

A portable optical bench was built to facilitate the alignment of the fibers in the injection module. One of the blue fibers has to overlap with one of the red fibers when projected back onto the sky, so that a star can be observed simultaneously in both wavelength ranges. We achieve this by back‐illuminating the fibers and observing the output from the injection module with a CCD camera and telephoto lens. The blue fiber connector is fixed with dowels, and the red fiber connector is manually adjusted until alignment is achieved. An overlap of ≥70% can be quickly achieved with this method.

In order not to lose starlight at the entrance of the fiber core, we can afford up to ±0.32 mm of defocus, which is far more than the 0.06 mm thermal deformation from + 20°C to −90°C of the chosen Invar‐Astrositall‐Pyrex combination. Ray‐tracing confirmed that the defocus between the red and blue images was also well within this limit (±0.18 mm around the best focus determined in white light) as both bundles of optical fibers are, by design, equidistant from the beamsplitter.

On the telescope side, two brass connectors have been machined to connect the fibers into the injection module. At the extremities of the fibers, the buffer is removed over a few millimeters so that the three claddings can be placed parallel and close together in a rectangular shaped groove. 5‐Minute Araldite epoxy adhesive is used to glue the fibers to the connectors under supervision using a magnifying eyepiece. The precision obtained was about ±20 μm in alignment and with a gap between the fibers of up to 30 μm. Final optical polishing of the connector end was performed to obtain efficient injection of light.

On the spectrograph side, a single brass connector is machined with six V‐shaped grooves separated by 0.8 mm. The same technique as described above is used to glue the fibers. The axes of the six fibers were made parallel with a precision better than 3 ^' (the angle between two fiber images on the CCD is 9 ^') by careful machining of the grooves.

We have determined the photon flux [N(λ) in photons cm⁻² s⁻¹ Å⁻¹] that should reach the CCD using the following assumptions (formula from Zombeck 1990, p. 103):

where T_e is the effective temperature (in K), E is the atmospheric extinction (we assume 0.4 at a zenith distance of 60° and at 500 nm), and M_b is the bolometric magnitude.

In the following calculation, we use Sirius (an A1 V star) with M_b = -1.69 and T_e = 9230 K. With the AFOS collecting surface of 660 cm², at λ = 500 nm and 60° zenith distance we expect 18.7 × 10⁶ photons s⁻¹ nm⁻¹. Knowing that the instrument transmission is approximately 4% at this wavelength and that the light is spatially spread over 4 pixels (each spectrum represents four lines on the CCD) with a spectral scale of 0.64 nm pixel⁻¹, we finally calculate a flux of 292,000 electrons s⁻¹ pixel⁻¹.

The spectra of Sirius in Figure 6 shows a flux of only about 62,000 electrons s⁻¹ pixel⁻¹ at 500 nm. The mismatch between observation and calculation could be accounted for by factors such as the quality of the injection into the fiber and will be investigated further. However, these unknown losses should not change in time, which should allow us to achieve meaningful photometry.

The minimum airglow intensity at visible wavelengths is typically 1 R (1 R = 10⁶ photons cm⁻² s⁻¹ sr⁻¹) or 6.8 × 10^-17 W m⁻² μm⁻¹ arcsec⁻², which is equivalent to a magnitude of 22 arcsec⁻². In a 20 ^'' fiber aperture, we expect to see about 2100 photons s⁻¹ (i.e., 5 electrons s⁻¹ pixel⁻¹).

This is still above the CCD dark current specified of 1 electron s⁻¹ pixel⁻¹ at −25°C, which permits observations limited by the sky background. Furthermore, after only 20 s of integration, the sky photon shot noise becomes dominant over the readout noise (10 electrons per pixel).